AI-Based Detection of Care Gaps in Electronic Health Records for Improved Clinical Decision Making Using Temporal, Semantic, and Guideline-Aware Learning

Abstract

Lapses in care (i.e., missed screenings, late follow-ups, missed medications, off-pathway deviations) remain a significant source of avoidable harm despite the widespread implementation of Electronic Health Records (EHRs) and clinical decision support systems based on rules (Clinical Decision Support, 2009). To address this issue, this manuscript proposes a guideline-aware artificial intelligence system that actively identifies gaps in care by combining structured EHR variables with unstructured clinical documents and tracking patient progress. Clinical guidelines are represented as machine-readable paths in a body of knowledge graph, which explicitly allow one to compare a patient's sequence of temporal events with the care actions desired for it. A hybrid structure integrates the BioBERT-based entity extraction functionality with a Transformer to longitudinally model and an attention-based fusion layer to generate a care-gap risk score with understandable explanations. Benchmark critical-care EHR data experiments expected results are high-precision of the system, a less-fatigued alert, and earlier detection of missed interventions as compared to rule-based CDSS and traditional machine-learned baselines. This designation takes clinical decision support to the next level by offering more than reactive alerts; it provides proactive, explainable detection of guideline violations, making it safer and more effective in providing quality care to patients.